Time horizon (METR)
The length of task, measured by how long a skilled human expert takes, at which an AI agent is predicted to succeed half the time.
METR is explicit that it is “a measure of the difficulty of a task, rather than the time an AI spends to complete the task.” It is not how long an agent can run on its own.
Read the 80% line too. For the top model on METR’s data, the 50% horizon is reported at 16 hours or more, but the 80% horizon is about 3 hours. METR warns that measurements above 16 hours are unreliable with its current task suite, which is mostly software engineering, machine learning and cybersecurity. The page was last updated on 8 September 2026 and is marked as no longer actively updated.
A statistical re-analysis of METR’s data (Nguyen and Fithian, 8 October 2026) finds that difficulty doesn’t scale evenly: going from 3 to 30 minutes is much easier than going from 30 minutes to 5 hours. Don’t draw a straight line to multi-day autonomy.
This model can work on its own for 16 hours.
On tasks a person needs 16 hours for, expect it to fail about half the time.
Plan unattended work on the 80% number, and check that the task type matches METR’s suite.
Use the trend for direction, not one number for promises.
Sources
- METR, Task-completion time horizons of frontier AI models · updated 8 Sep 2026
- Nguyen and Fithian, On the estimation and validity of AI time horizons · 8 Oct 2026
- METR (summary of published estimates), Wikipedia · Oct 2026